A Demand-aware Networked System Using Telemetry and ML with ReactNET

TL;DR

ReactNet addresses the challenge of delivering QoS/QoE guarantees in dynamic, shared networks by integrating a programmable data plane with machine learning-based traffic classification. The architecture collects fine-grained telemetry via P4, labels data in the control plane, trains ML classifiers, and adjusts forwarding rules through a feedback loop to meet application QoS goals. Proof-of-concept experiments in Mininet with a video-streaming workload show improved QoE metrics, while IoT traces demonstrate high classifier accuracy (up to 99%) across different ML models. This work demonstrates the feasibility of demand-aware, self-adjusting networks and suggests directions for richer on-switch ML integration and more advanced adaptation policies.

Abstract

Emerging network applications ranging from video streaming to virtual/augmented reality need to provide stringent quality-of-service (QoS) guarantees in complex and dynamic environments with shared resources. A promising approach to meeting these requirements is to automate complex network operations and create self-adjusting networks. These networks should automatically gather contextual information, analyze how to efficiently ensure QoS requirements, and adapt accordingly. This paper presents ReactNET, a self-adjusting networked system designed to achieve this vision by leveraging emerging network programmability and machine learning techniques. Programmability empowers ReactNET by providing fine-grained telemetry information, while machine learning-based classification techniques enable the system to learn and adjust the network to changing conditions. Our preliminary implementation of ReactNET in P4 and Python demonstrates its effectiveness in video streaming applications.

Figures (7)

• Figure 1: The architecture of ReactNet for self-adjustable networks, implemented on a P4 programmable switch. The data for training, i.e., packet headers and fine-grained telemetry information, of the system is collected via the data plane, while the classification of packets ( to three classes in this figure), optimizing, and updating the network are performed via the control plane.
• Figure 2: Our evaluation network topology for the video streaming application.
• Figure 3: The effect of ReactNet on the PSNR metric of the streamed video.
• Figure 4: The effect of ReactNet on the SSIM metric of the streamed video.
• Figure 5: The impact of using ReactNet on the VMAF metric of the streamed video.